GPT-6 Astra vs Kimi K3
Wins 6 of 8 areas
Agents · Reasoning · Facts · Images and charts · Math · Following instructions
Wins 2 of 8 areas
Coding · Long documents
GPT-6 Astra is the stronger all-rounder.Kimi K3 is cheaper and better at coding.
Both rank among the ten best models we track in agents.
Scores updated · 39 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- ReasoningHard problems that need careful thinking60GPT-6 Astra6 of 6 tests
- FactsGetting facts right instead of making them up30GPT-6 Astra3 of 3 tests
- MathCompetition and research-level math30GPT-6 Astra3 of 3 tests
- AgentsCarrying out multi-step tasks on its own21GPT-6 Astra2 of 5 tests · 2 ties
- Images and chartsUnderstanding pictures, charts and video10GPT-6 Astra1 of 1 test
- Following instructionsDoing exactly what it is asked10GPT-6 Astra1 of 1 test
- CodingWriting and fixing software23Kimi K33 of 5 tests
- Long documentsFinding answers in very long texts01Kimi K31 of 1 test
Images and charts, long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Kimi K3 costs 70% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-6 Astra pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+58.6points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+46.5points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+34.6points ahead
Where Kimi K3 pulls ahead
- Reasons across sets of long documentsAA-LCR+8points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+4.9points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+4.6points ahead
Every test, side by side
All 39 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K3
- LMArena · WebDevGPT-6 Astra by 120 rating points17751655+120 rating
- LiveBench · Agentic CodingKimi K3 by 4.957.362.2+4.9
- Terminal-Bench 2.1GPT-6 Astra by 3.488.485+3.4
- SciCodeKimi K3 by 356.559.5+3
- LiveBench · CodingKimi K3 by 1.180.481.5+1.1
AgentsGPT-6 Astra
- Terminal-Bench 4.0GPT-6 Astra by 46.559.112.6+46.5
- τ-Bench V3 · BankingKimi K3 by 4.641.446+4.6
- GDPValGPT-6 Astra by 1.953.651.7+1.9
- AA IT-Bench SREtie48.647.7tie
- BrowseComptie91.591.2tie
ReasoningGPT-6 Astra
- ARC-AGI-2GPT-6 Astra by 34.69560.4+34.6
- SimpleBenchGPT-6 Astra by 22.983.660.7+22.9
- CritPtGPT-6 Astra by 8.331.723.4+8.3
- Humanity's Last ExamGPT-6 Astra by 7.854.746.9+7.8
- GPQA DiamondGPT-6 Astra by 2.696.193.5+2.6
- LiveBench · ReasoningGPT-6 Astra by 292.790.7+2
FactsGPT-6 Astra
- SimpleQA VerifiedGPT-6 Astra by 2575.650.6+25
- AA-Omniscience · AccuracyGPT-6 Astra by 1562.647.6+15
- AA-Omniscience · Non-hallucinationGPT-6 Astra by 1.948.746.8+1.9
MathGPT-6 Astra
- FrontierMath Tier 4GPT-6 Astra by 58.697.639+58.6
- FrontierMath Tiers 1-3 (v2)GPT-6 Astra by 21.593.772.2+21.5
- LiveBench · MathematicsGPT-6 Astra by 12.496.884.4+12.4
Following instructionsGPT-6 Astra
- LiveBench · Instruction FollowingGPT-6 Astra by 4.275.671.4+4.2
Other results14 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- Terminal-Bench 4.0GPT-6 Astra by 45.157.712.6+45.1
- Agents' Last ExamGPT-6 Astra by 3159.328.3+31
- AA-OmniscienceGPT-6 Astra by 23.743.419.7+23.7
- HiL-Bench (Tools-allowed)GPT-6 Astra by 22.665.342.7+22.6
- OSWorld 2.0GPT-6 Astra by 14.372.658.3+14.3
- AA IntelligenceGPT-6 Astra by 9.152.743.6+9.1
- PostTrainBenchGPT-6 Astra by 7.744.336.6+7.7
- DeepSWE 1.1GPT-6 Astra by 5.174.169+5.1
- livebench_data_analysisGPT-6 Astra by 4.38378.7+4.3
- livebench_languageGPT-6 Astra by 3.989.485.5+3.9
- ARC-AGI-1GPT-6 Astra by 397.594.5+3
- HLE (with tools)Kimi K3 by 2.657.259.8+2.6
- AA Agentic Indextie51.550.6tie
- Artificial Analysis Coding Indextie76.976.2tie
Questions people ask
Which is better, GPT-6 Astra or Kimi K3?
GPT-6 Astra wins six of the eight areas we test: agents, reasoning, facts, images and charts, math and following instructions. Kimi K3 wins coding and long documents, and costs 70% less.
Which is better for coding?
Kimi K3. It wins 3 of the 5 coding tests both models report; GPT-6 Astra wins 2.
Which is cheaper?
GPT-6 Astra costs $10.00 per million input tokens and $50.00 per million output tokens; Kimi K3 costs $3.00 and $15.00. That makes Kimi K3 about 70% cheaper for the same work.
How do you compare the two?
We use the 39 benchmark tests both models have published scores on. The verdict counts the 25 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 14 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.